speechflow 1.2.1 → 1.2.3

package/src/speechflow-node-a2a-vad.ts

@@ -47,12 +47,13 @@ export default class SpeechFlowNodeVAD extends SpeechFlowNode {
 
         /*  declare node configuration parameters  */
         this.configure({
-            mode:               { type: "string", val: "unplugged", match: /^(?:silenced|unplugged)$/ },
+            mode:               { type: "string", val: "silenced", match: /^(?:silenced|unplugged)$/ },
             posSpeechThreshold: { type: "number", val: 0.50 },
             negSpeechThreshold: { type: "number", val: 0.35 },
             minSpeechFrames:    { type: "number", val: 2    },
             redemptionFrames:   { type: "number", val: 12   },
-            preSpeechPadFrames: { type: "number", val: 1    }
+            preSpeechPadFrames: { type: "number", val: 1    },
+            postSpeechTail:     { type: "number", val: 1500 }
         })
 
         /*  declare node input/output format  */
@@ -74,6 +75,8 @@ export default class SpeechFlowNodeVAD extends SpeechFlowNode {
         const vadSamplesPerFrame  = 512   /* required for VAD v5 */
 
         /*  establish Voice Activity Detection (VAD) facility  */
+        let tail = false
+        let tailTimer: ReturnType<typeof setTimeout> | null = null
         this.vad = await RealTimeVAD.new({
             model:                   "v5",
             sampleRate:              this.config.audioSampleRate, /* before resampling to 16KHz */
@@ -85,13 +88,38 @@ export default class SpeechFlowNodeVAD extends SpeechFlowNode {
             preSpeechPadFrames:      this.params.preSpeechPadFrames,
             onSpeechStart: () => {
                 log("info", "VAD: speech start")
+                if (this.params.mode === "unlugged") {
+                    tail = false
+                    if (tailTimer !== null) {
+                        clearTimeout(tailTimer)
+                        tailTimer = null
+                    }
+                }
             },
             onSpeechEnd: (audio) => {
                 const duration = utils.audioArrayDuration(audio, vadSampleRateTarget)
                 log("info", `VAD: speech end (duration: ${duration.toFixed(2)}s)`)
+                if (this.params.mode === "unlugged") {
+                    tail = true
+                    if (tailTimer !== null)
+                        clearTimeout(tailTimer)
+                    tailTimer = setTimeout(() => {
+                        tail = false
+                        tailTimer = null
+                    }, this.params.postSpeechTail)
+                }
             },
             onVADMisfire: () => {
                 log("info", "VAD: speech end (segment too short)")
+                if (this.params.mode === "unlugged") {
+                    tail = true
+                    if (tailTimer !== null)
+                        clearTimeout(tailTimer)
+                    tailTimer = setTimeout(() => {
+                        tail = false
+                        tailTimer = null
+                    }, this.params.postSpeechTail)
+                }
             },
             onFrameProcessed: (audio) => {
                 /*  annotate the current audio segment  */
@@ -99,7 +127,7 @@ export default class SpeechFlowNodeVAD extends SpeechFlowNode {
                 if (element === undefined || element.type !== "audio-frame")
                     throw new Error("internal error which cannot happen: no more queued element")
                 const segment = element.segmentData[element.segmentIdx++]
-                segment.isSpeech = (audio.isSpeech > audio.notSpeech)
+                segment.isSpeech = (audio.isSpeech > audio.notSpeech) || tail
 
                 /*  annotate the entire audio chunk  */
                 if (element.segmentIdx >= element.segmentData.length) {
@@ -124,6 +152,7 @@ export default class SpeechFlowNodeVAD extends SpeechFlowNode {
             writableObjectMode: true,
             readableObjectMode: true,
             decodeStrings:      false,
+            highWaterMark:      1,
 
             /*  receive audio chunk (writable side of stream)  */
             write (chunk: SpeechFlowChunk, encoding, callback) {

package/src/speechflow-node-a2a-wav.ts

@@ -101,6 +101,7 @@ export default class SpeechFlowNodeWAV extends SpeechFlowNode {
             readableObjectMode: true,
             writableObjectMode: true,
             decodeStrings:      false,
+            highWaterMark:      1,
             transform (chunk: SpeechFlowChunk, encoding, callback) {
                 if (!Buffer.isBuffer(chunk.payload))
                     callback(new Error("invalid chunk payload type"))

package/src/speechflow-node-a2t-deepgram.ts

@@ -180,6 +180,7 @@ export default class SpeechFlowNodeDeepgram extends SpeechFlowNode {
             writableObjectMode: true,
             readableObjectMode: true,
             decodeStrings:      false,
+            highWaterMark:      1,
             write (chunk: SpeechFlowChunk, encoding, callback) {
                 if (chunk.type !== "audio")
                     callback(new Error("expected audio input chunk"))

package/src/speechflow-node-t2a-elevenlabs.ts

@@ -128,6 +128,7 @@ export default class SpeechFlowNodeElevenlabs extends SpeechFlowNode {
             writableObjectMode: true,
             readableObjectMode: true,
             decodeStrings:      false,
+            highWaterMark:      1,
             transform (chunk: SpeechFlowChunk, encoding, callback) {
                 if (Buffer.isBuffer(chunk.payload))
                     callback(new Error("invalid chunk payload type"))

package/src/speechflow-node-t2a-kokoro.ts

@@ -121,6 +121,7 @@ export default class SpeechFlowNodeKokoro extends SpeechFlowNode {
             writableObjectMode: true,
             readableObjectMode: true,
             decodeStrings:      false,
+            highWaterMark:      1,
             transform (chunk: SpeechFlowChunk, encoding, callback) {
                 if (Buffer.isBuffer(chunk.payload))
                     callback(new Error("invalid chunk payload type"))

package/src/speechflow-node-t2t-deepl.ts

@@ -74,6 +74,7 @@ export default class SpeechFlowNodeDeepL extends SpeechFlowNode {
             readableObjectMode: true,
             writableObjectMode: true,
             decodeStrings:      false,
+            highWaterMark:      1,
             transform (chunk: SpeechFlowChunk, encoding, callback) {
                 if (Buffer.isBuffer(chunk.payload))
                     callback(new Error("invalid chunk payload type"))

package/src/speechflow-node-t2t-format.ts

@@ -46,6 +46,7 @@ export default class SpeechFlowNodeFormat extends SpeechFlowNode {
             readableObjectMode: true,
             writableObjectMode: true,
             decodeStrings:      false,
+            highWaterMark:      1,
             transform (chunk: SpeechFlowChunk, encoding, callback) {
                 if (Buffer.isBuffer(chunk.payload))
                     callback(new Error("invalid chunk payload type"))

package/src/speechflow-node-t2t-ollama.ts

@@ -222,6 +222,7 @@ export default class SpeechFlowNodeOllama extends SpeechFlowNode {
             readableObjectMode: true,
             writableObjectMode: true,
             decodeStrings:      false,
+            highWaterMark:      1,
             transform (chunk: SpeechFlowChunk, encoding, callback) {
                 if (Buffer.isBuffer(chunk.payload))
                     callback(new Error("invalid chunk payload type"))

package/src/speechflow-node-t2t-openai.ts

@@ -203,6 +203,7 @@ export default class SpeechFlowNodeOpenAI extends SpeechFlowNode {
             readableObjectMode: true,
             writableObjectMode: true,
             decodeStrings:      false,
+            highWaterMark:      1,
             transform (chunk: SpeechFlowChunk, encoding, callback) {
                 if (Buffer.isBuffer(chunk.payload))
                     callback(new Error("invalid chunk payload type"))

package/src/speechflow-node-t2t-subtitle.ts

@@ -63,6 +63,7 @@ export default class SpeechFlowNodeSubtitle extends SpeechFlowNode {
             readableObjectMode: true,
             writableObjectMode: true,
             decodeStrings:      false,
+            highWaterMark:      1,
             transform (chunk: SpeechFlowChunk, encoding, callback) {
                 if (Buffer.isBuffer(chunk.payload))
                     callback(new Error("invalid chunk payload type"))

package/src/speechflow-node-t2t-transformers.ts

@@ -200,6 +200,7 @@ export default class SpeechFlowNodeTransformers extends SpeechFlowNode {
             readableObjectMode: true,
             writableObjectMode: true,
             decodeStrings:      false,
+            highWaterMark:      1,
             transform (chunk: SpeechFlowChunk, encoding, callback) {
                 if (Buffer.isBuffer(chunk.payload))
                     callback(new Error("invalid chunk payload type"))

package/src/speechflow-node-x2x-filter.ts

@@ -86,12 +86,13 @@ export default class SpeechFlowNodeFilter extends SpeechFlowNode {
             writableObjectMode: true,
             readableObjectMode: true,
             decodeStrings:      false,
+            highWaterMark:      1,
             transform (chunk: SpeechFlowChunk, encoding, callback) {
                 let val1: any
                 const val2: any = self.params.val
                 const m = self.params.var.match(/^meta:(.+)$/)
                 if (m !== null)
-                    val1 = chunk.meta.get(m[1])
+                    val1 = chunk.meta.get(m[1]) ?? ""
                 else if (self.params.key === "payload:length")
                     val1 = chunk.payload.length
                 else if (self.params.key === "payload:text")
@@ -100,7 +101,7 @@ export default class SpeechFlowNodeFilter extends SpeechFlowNode {
                     val1 = chunk.timestampStart.toMillis()
                 else if (self.params.key === "time:end")
                     val1 = chunk.timestampEnd.toMillis()
-                if (comparison(val1, self.params.ops, val2))
+                if (comparison(val1, self.params.op, val2))
                     this.push(chunk)
                 callback()
             },

package/src/speechflow-node-x2x-trace.ts

@@ -49,6 +49,7 @@ export default class SpeechFlowNodeTrace extends SpeechFlowNode {
             writableObjectMode: true,
             readableObjectMode: true,
             decodeStrings:      false,
+            highWaterMark:      1,
             transform (chunk: SpeechFlowChunk, encoding, callback) {
                 let error: Error | undefined
                 const fmtTime = (t: Duration) => t.toFormat("hh:mm:ss.SSS")
@@ -74,12 +75,12 @@ export default class SpeechFlowNodeTrace extends SpeechFlowNode {
                 }
                 else {
                     if (type === "text")
-                        log("debug", `${type} chunk: type=${chunk.type}` +
+                        log("debug", `${type} chunk: type=${chunk.type} ` +
                             `kind=${chunk.kind} ` +
                             `start=${fmtTime(chunk.timestampStart)} ` +
                             `end=${fmtTime(chunk.timestampEnd)} ` +
                             `payload-type=String payload-length=${chunk.payload.length} ` +
-                            `payload-encoding=${encoding} payload-content="${chunk.payload.toString()}" ` +
+                            `payload-content="${chunk.payload.toString()}" ` +
                             `meta=${fmtMeta(chunk.meta)}`)
                     else
                         error = new Error(`${type} chunk: seen String instead of Buffer chunk type`)

package/src/speechflow-node-xio-device.ts

@@ -199,6 +199,7 @@ export default class SpeechFlowNodeDevice extends SpeechFlowNode {
     async close () {
         /*  shutdown PortAudio  */
         if (this.io !== null) {
+            this.io.abort()
             this.io.quit()
             this.io = null
         }

package/src/speechflow-node-xio-mqtt.ts

@@ -111,6 +111,7 @@ export default class SpeechFlowNodeMQTT extends SpeechFlowNode {
             writableObjectMode: true,
             readableObjectMode: true,
             decodeStrings:      false,
+            highWaterMark:      1,
             write (chunk: SpeechFlowChunk, encoding, callback) {
                 if (mode === "r")
                     callback(new Error("write operation on read-only node"))

package/src/speechflow-node-xio-websocket.ts

@@ -114,6 +114,7 @@ export default class SpeechFlowNodeWebsocket extends SpeechFlowNode {
                 writableObjectMode: true,
                 readableObjectMode: true,
                 decodeStrings:      false,
+                highWaterMark:      1,
                 write (chunk: SpeechFlowChunk, encoding, callback) {
                     if (mode === "r")
                         callback(new Error("write operation on read-only node"))
@@ -195,6 +196,7 @@ export default class SpeechFlowNodeWebsocket extends SpeechFlowNode {
                 writableObjectMode: true,
                 readableObjectMode: true,
                 decodeStrings:      false,
+                highWaterMark:      1,
                 write (chunk: SpeechFlowChunk, encoding, callback) {
                     if (mode === "r")
                         callback(new Error("write operation on read-only node"))

package/src/speechflow-utils.ts

@@ -68,6 +68,7 @@ export function createTransformStreamForWritableSide () {
         readableObjectMode: true,
         writableObjectMode: true,
         decodeStrings: false,
+        highWaterMark: 1,
         transform (chunk: SpeechFlowChunk, encoding, callback) {
             this.push(chunk.payload)
             callback()
@@ -86,6 +87,7 @@ export function createTransformStreamForReadableSide (type: "text" | "audio", ge
         readableObjectMode: true,
         writableObjectMode: true,
         decodeStrings: false,
+        highWaterMark: (type === "audio" ? 19200 : 65536),
         transform (chunk: Buffer | string, encoding, callback) {
             const timeZero = getTimeZero()
             const start = DateTime.now().diff(timeZero)

package/src/speechflow.ts

@@ -61,11 +61,11 @@ type wsPeerInfo = {
             "Usage: $0 " +
             "[-h|--help] " +
             "[-V|--version] " +
+            "[-S|--status] " +
             "[-v|--verbose <level>] " +
             "[-a|--address <ip-address>] " +
             "[-p|--port <tcp-port>] " +
             "[-C|--cache <directory>] " +
-            "[-S|--status] " +
             "[-e|--expression <expression>] " +
             "[-f|--file <file>] " +
             "[-c|--config <id>@<yaml-config-file>] " +
@@ -80,6 +80,14 @@ type wsPeerInfo = {
             default:  false,
             describe: "show program version information"
         })
+        .option("S", {
+            alias:    "status",
+            type:     "boolean",
+            array:    false,
+            coerce,
+            default:  false,
+            describe: "show one-time status of nodes"
+        })
         .option("v", {
             alias:    "log-level",
             type:     "string",
@@ -116,14 +124,6 @@ type wsPeerInfo = {
             default:  path.join(dataDir, "cache"),
             describe: "directory for cached files (primarily AI model files)"
         })
-        .option("S", {
-            alias:    "status",
-            type:     "boolean",
-            array:    false,
-            coerce,
-            default:  false,
-            describe: "show one-time status of nodes"
-        })
         .option("e", {
             alias:    "expression",
             type:     "string",
@@ -423,6 +423,7 @@ type wsPeerInfo = {
     }
 
     /*  graph processing: PASS 3: open nodes  */
+    const timeZero = DateTime.now()
     for (const node of graphNodes) {
         /*  connect node events  */
         node.on("log", (level: string, msg: string, data?: any) => {
@@ -434,20 +435,14 @@ type wsPeerInfo = {
 
         /*  open node  */
         cli!.log("info", `open node <${node.id}>`)
+        node.setTimeZero(timeZero)
         await node.open().catch((err: Error) => {
             cli!.log("error", `[${node.id}]: ${err.message}`)
             throw new Error(`failed to open node <${node.id}>`)
         })
     }
 
-    /*  graph processing: PASS 4: set time zero in all nodes  */
-    const timeZero = DateTime.now()
-    for (const node of graphNodes) {
-        cli!.log("info", `set time zero in node <${node.id}>`)
-        node.setTimeZero(timeZero)
-    }
-
-    /*  graph processing: PASS 5: connect node streams  */
+    /*  graph processing: PASS 4: connect node streams  */
     for (const node of graphNodes) {
         if (node.stream === null)
             throw new Error(`stream of node <${node.id}> still not initialized`)